On Fri, Jan 08, 2010 at 11:54:21AM -0600, Shawn Walker wrote: > On 01/ 7/10 06:55 PM, Brock Pytlik wrote: > >Dan noticed that a certain search test seemed to be stalling out during > >testing, after some further investigation, it appears that getting the > >versions is the location of the stall because of a slow data rate from > >the server. > > > >At this point, I'm not sure where to go next to figure out what's going > >on, so I thought I'd send what I've gathered so far out. This is > >reproducible on my and Dan's machine by doing: > >pfexec tests/run.py -o 2989 -t -v > > > >Disabling persistent depot removes the stall. The stall causes a test > >that when run alone, or with no persistent depot to take 4 seconds now > >takes about 60. It looks to me like the first call to versions gets the > >Exception, and the second almost does, taking another 27 seconds to read > >the result from versions/0. > > For those wondering, the result of further research into this issue > resulted in this bug: > > http://defect.opensolaris.org/bz/show_bug.cgi?id=13746
Well, the root cause was that the test suite wasn't garbage collecting orphaned API/Image/Transport instances. The transport's connections remained open, pinning down a relatively sparse number of depot worker threads. Calling gc.collect() in the test suite caused the problem to vanish. However, since we thought that it might actually be useful to explicitly shutdown the transport once we finished a download, we're fixing the problem in the client and transport code instead. -j _______________________________________________ pkg-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/pkg-discuss
